akino222333/dobot_Xtrainer_teddybear_pickup_passover_and_place_dataset

---
pretty_name: Robot Multi-View HDF5 Dataset
license: mit
task_categories:
- robotics
- computer-vision
- imitation-learning
---

# Robot Multi-View HDF5 Dataset

## Dataset description
This dataset was converted from episode-level HDF5 files into a frame-level Hugging Face dataset.

- Total episode files: 100
- Split: train only
- Each row corresponds to one frame
- Each row contains:
- left wrist camera image
- right wrist camera image
- top camera image
- qpos
- qvel
- action

## Data structure

Columns:
- `episode_id`: integer episode index after sorting the input HDF5 files
- `episode_file`: original HDF5 file name
- `frame_id`: frame index inside the episode
- `left_wrist`: left wrist RGB image
- `right_wrist`: right wrist RGB image
- `top`: top RGB image
- `qpos`: original 14D robot state vector
- `qvel`: original 14D robot velocity/state vector
- `action`: original 14D action vector

## Notes
This dataset preserves the original 14D definitions of `qpos`, `qvel`, and `action` without splitting them into subfields.

Users can perform their own train/test split after downloading the dataset.

## Loading example

```python
from datasets import load_dataset
ds = load_dataset("akino222333/dobot_Xtrainer_teddybear_pickup_passover_and_place_dataset")
print(ds["train"][0].keys())
```